Automate your operations before you automate your marketing
Most businesses reach for AI in the place where mistakes are cheapest and value is lowest. The money is in the boring pipeline nobody writes about.
There is a predictable order in which businesses adopt AI, and it is almost exactly backwards.
They start with marketing copy. Then customer support replies. Then, if the enthusiasm survives, some kind of internal chatbot that nobody uses twice. Meanwhile the actual machinery of the business — the pipeline that takes an order and turns it into a delivered thing — runs on the same manual steps it ran on in 2015.
I understand why. Marketing is where AI is most visible, most demonstrable, and least likely to break anything if it goes wrong. A bad blog post costs nothing. A bad dispatch decision costs a customer.
But that reasoning contains its own refutation. The reason automating operations feels dangerous is that operations is where the value is.
The asymmetry nobody mentions
Marketing automation competes against a task that was already cheap. Writing a product description takes ten minutes and any competent person can do it. Automating it saves ten minutes and produces something slightly worse.
Operations automation competes against a task that is expensive precisely because it requires judgement, context, and someone who has been there three years. When you automate a step in a production pipeline, you are not saving ten minutes. You are removing a bottleneck that gates everything downstream of it, on every job, every day, forever.
Those are different orders of magnitude, and the second one compounds.
Why almost nobody does it
Because it is genuinely hard, and because the failures are visible.
An operations automation has to be correct. Not usually correct — correct, or correct with a clean escalation path to a human who can catch the rest. It has to handle the file that arrives malformed, the order with the impossible delivery date, the customer who changed their mind after the job hit the floor. Every one of those edge cases is a real event that happened to a real business, and any automation that ignores them will fail in front of a customer within a fortnight.
Marketing content has no such constraint. If it is mediocre, nothing happens. That absence of consequence is exactly what makes it a poor place to spend your automation budget — and, incidentally, why the internet is now full of writing that nothing depended on.
What this site is for
I run automation inside a print and design manufacturing business, which is an unglamorous place to work on this problem and therefore an unusually honest one. Files arrive broken. Jobs move across a physical floor. Carriers lose packages. There is no version of this pipeline where an automation can be approximately right and nobody notices.
Everything I publish here comes out of systems that had to survive that. Where I have numbers, I will give the numbers. Where an automation failed, I will describe how, because the failure modes are the part that transfers between businesses and the successes usually do not.
I have no course to sell you and no tool to affiliate. I am writing this because when I was building these systems, what I wanted to read did not exist. Everyone was describing the demo. Nobody was describing the Monday morning three months in, when the intervention rate creeps up and you have to work out whether you have built an automation or a second job.
That is the writing I am going to try to do.